Analysing the feasibility of an automated AI-based classifier for detecting paranasal anomalies in the maxillary sinus

Introduction  Large scale population studies have been performed to analyse the rate of finding sinus opacities in cranial MRIs. It is of interest whether there are findings requiring clarification. Using AI-based methods can automate the detection of the sinus opacities and reduce the workload of clinicians. In this work, a method for AI-based classification was developed in order to automatically recognise paranasal sinus opacities.

Methods As part of the Hamburg City Health Study (HCHS), cMRIs of participants (45-74 years) were recorded for neuroradiological assessment. The following questions were addressed: 1. Is there an opacity of the maxillary sinus: yes/no? 2.Differentiation of opacity: mucosal thickening, polyp(s) or cyst(s). All MRIs (199) were annotated by specialists and the results of the AI were compared with this as a "gold standard". 106 participants showed inconspicuous and 93 participants maxillary sinuses with opacities. The AI-based classification system was carried out by a neural network (3D ResNet18), the data analysis was performed using a 5-fold cross-validation.

Results Considering “opacity" as the positive class, our AI classification system showed high classification accuracy (F1 score of 0.70±0.06 and an Area under Receiver Operating Characteristic (AuROC) of 0.85±0.03). In the further classification of opacities, our AI-based method achieved an accuracy of 100% for polyps, 60% for cysts and 45% for mucosal thickening.

Conclusions Our feasibility analysis shows a useful application for AI-based classification of the maxillary sinuses and can support radiological sinus findings.

Publication History

Article published online:
12 May 2023